Pith. Sign in

RECORD

REPT: Bridging Language Models and Machine Reading Comprehension via Retrieval-Based Pre-training

PaperPDF from arXiv

data files

Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Fangkai Jiao, Yangyang Guo, Yilin Niu, Feng Ji, Feng-Lin Li, Liqiang Nie

Computation and Language · Submitted 10 May 2021

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim Experimental results demonstrate the effectiveness of our pre-training approach.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2105.04201 (version 2)
Pith IDZJHV77J4, Pith's permanent record of this paper and its review
CitationFangkai Jiao, Feng Ji, Feng-Lin Li, Liqiang Nie, Yangyang Guo, Yilin Niu. "REPT: Bridging Language Models and Machine Reading Comprehension via Retrieval-Based Pre-training." arXiv:2105.04201 (2021). https://pith.science/paper/ZJHV77J4
Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 5,533 input tokens.

What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.15 · human participants 0.03 · controlled comparison 0.19 · randomized assignment 0.01 · states quantitative result 0.13 · compares with prior results 0.20 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; .

Pre-trained Language Models (PLMs) have achieved great success on Machine Reading Comprehension (MRC) over the past few years. Although the general language representation learned from large-scale corpora does benefit MRC, the poor support in evidence extraction which requires reasoning across multiple sentences hinders PLMs from further advancing MRC. To bridge the gap between general PLMs and MRC, we present REPT, a REtrieval-based Pre-Training approach. In particular, we introduce two self-supervised tasks to strengthen evidence extraction during pre-training, which is further inherited by downstream MRC tasks through the consistent retrieval operation and model architecture. To evaluate our proposed method, we conduct extensive experiments on five MRC datasets that require collecting evidence from and reasoning across multiple sentences. Experimental results demonstrate the effectiveness of our pre-training approach. Moreover, further analysis shows that our approach is able to enhance the capacity of evidence extraction without explicit supervision.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 Pre-trained Language Models (PLMs) have achieved great success on Machine Reading Comprehension (MRC) over the past few years. background 1.00 0.04 0.02 asserted 1.00
s1 Although the general language representation learned from large-scale corpora does benefit MRC, the poor support in evidence extraction which requires reasoning across multiple sentences hinders PLMs from further advancing MRC. gap 0.72 0.12 0.04 asserted 1.00
s2 To bridge the gap between general PLMs and MRC, we present REPT, a REtrieval-based Pre-Training approach. contribution 0.95 0.91 0.07 asserted 0.96
s3 In particular, we introduce two self-supervised tasks to strengthen evidence extraction during pre-training, which is further inherited by downstream MRC tasks through the consistent retrieval operation and model architecture. method 0.89 0.92 0.07 asserted 1.00
s4 To evaluate our proposed method, we conduct extensive experiments on five MRC datasets that require collecting evidence from and reasoning across multiple sentences. method 0.93 0.96 0.03 asserted 0.93
s5 Experimental results demonstrate the effectiveness of our pre-training approach. result 1.00 0.92 0.65 asserted 1.00
s6 Moreover, further analysis shows that our approach is able to enhance the capacity of evidence extraction without explicit supervision. result 0.55 0.95 0.27 asserted 0.96

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v2
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper
Original abstract

Pre-trained Language Models (PLMs) have achieved great success on Machine Reading Comprehension (MRC) over the past few years. Although the general language representation learned from large-scale corpora does benefit MRC, the poor support in evidence extraction which requires reasoning across multiple sentences hinders PLMs from further advancing MRC. To bridge the gap between general PLMs and MRC, we present REPT, a REtrieval-based Pre-Training approach. In particular, we introduce two self-supervised tasks to strengthen evidence extraction during pre-training, which is further inherited by downstream MRC tasks through the consistent retrieval operation and model architecture. To evaluate our proposed method, we conduct extensive experiments on five MRC datasets that require collecting evidence from and reasoning across multiple sentences. Experimental results demonstrate the effectiveness of our pre-training approach. Moreover, further analysis shows that our approach is able to enhance the capacity of evidence extraction without explicit supervision.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

What this paper does with the arXiv papers it cites

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

What it cites · 2 sentences without a confident label

Uses the method or tool of 3 papers
  • We adopt a 2-layer feed-forward network with GeLU activation [ref] and layer normalization [ref] to predict the masked entities and nouns.

    Gaussian Error Linear Units (GELUs) · Normalized Feed-forward Network · 0.99

  • We adopt a 2-layer feed-forward network with GeLU activation [ref] and layer normalization [ref] to predict the masked entities and nouns.

    Layer Normalization · Normalized Feed-forward Network · 0.99

  • We built our model on Huggingface's Pytorch transformer repository [ref] , and used AdamW [ref] as the optimizer.

    HuggingFace's Transformers: State-of-the-art Natural Language Processing · Implementation Detail · 1.00

Cites as support 1 paper
Cites as background 2 papers

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.